Hsuxu/pytorch_warmup

Learning Rate Warmup in PyTorch

A PyTorch Extension for Learning Rate Warmup

This library contains PyTorch implementations of the warmup schedules described in On the adequacy of untuned warmup for adaptive optimization.

Installation

Make sure you have Python 3.6+ and PyTorch 1.1+. Then, run the following command:

python setup.py install

or

pip install -U pytorch_warmup

Usage

Sample Codes

The scheduled learning rate is dampened by the multiplication of the warmup factor:

Approach 1

When the learning rate schedule uses the global iteration number, the untuned linear warmup can be used as follows:

import torch
import pytorch_warmup as warmup

optimizer = torch.optim.AdamW(params, lr=0.001, betas=(0.9, 0.999), weight_decay=0.01)
num_steps = len(dataloader) * num_epochs
lr_scheduler = torch.optim.lr_scheduler.CosineAnnealingLR(optimizer, T_max=num_steps)
warmup_scheduler = warmup.UntunedLinearWarmup(optimizer)
for epoch in range(1,num_epochs+1):
    for batch in dataloader:
        optimizer.zero_grad()
        loss = ...
        loss.backward()
        optimizer.step()
        lr_scheduler.step()
        warmup_scheduler.dampen()

For PyTorch 1.4 or above, use an LR scheduler as the following:

        lr_scheduler.step(lr_scheduler.last_epoch+1)

Approach 2

When the learning rate schedule uses the epoch number, the warmup schedule can be used as follows (for PyTorch 1.2 or above):

lr_scheduler = torch.optim.lr_scheduler.MultiStepLR(optimizer, milestones=[num_epochs//3], gamma=0.1)
warmup_scheduler = warmup.UntunedLinearWarmup(optimizer)
warmup_scheduler.last_step = -1 # initialize the step counter
for epoch in range(1,num_epochs+1):
    for batch in dataloader:
        lr_scheduler.step(epoch-1)
        warmup_scheduler.dampen()
        optimizer.zero_grad()
        loss = ...
        loss.backward()
        optimizer.step()

The user warning about calling lr_scheduler.step() before optimizer.step() may be ignored.

Warmup Schedules

Manual Warmup

The warmup factor w(t) depends on the warmup period, which must manually be specified, for LinearWarmup and ExponentialWarmup.

Linear

w(t) = min(1, t / warmup_period)

warmup_scheduler = warmup.LinearWarmup(optimizer, warmup_period=2000)

Exponential

w(t) = 1 - exp(-t / warmup_period)

warmup_scheduler = warmup.ExponentialWarmup(optimizer, warmup_period=1000)

Untuned Warmup

The warmup period is given by a function of Adam's beta2 parameter for UntunedLinearWarmup and UntunedExponentialWarmup.

Linear

warmup_period = 2 / (1 - beta2)

warmup_scheduler = warmup.UntunedLinearWarmup(optimizer)

Exponential

warmup_period = 1 / (1 - beta2)

warmup_scheduler = warmup.UntunedExponentialWarmup(optimizer)

RAdam Warmup

The warmup factor depends on Adam's beta2 parameter for RAdamWarmup. Please see the original paper for the details.

warmup_scheduler = warmup.RAdamWarmup(optimizer)

Apex's Adam

The Apex library provides an Adam optimizer tuned for CUDA devices, FusedAdam. The FusedAdam optimizer can be used with the warmup schedulers. For example:

optimizer = apex.optimizers.FusedAdam(params, lr=0.001, betas=(0.9, 0.999), weight_decay=0.01)
lr_scheduler = torch.optim.lr_scheduler.CosineAnnealingLR(optimizer, T_max=num_steps)
warmup_scheduler = warmup.UntunedLinearWarmup(optimizer)

License

MIT License

Copyright (c) 2019 Takenori Yamamoto